tangledlabs/tangled-llama-pints-1.5b-v0.2-instruct

See axolotl config

axolotl version: 0.4.1

base_model: pints-ai/1.5-Pints-16K-v0.1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: tangledgroup/tangled-llama-pints-1.5b-v0.2-dataset
    type: sharegpt
    conversation: chatml
chat_template: chatml
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/qlora-out

adapter: qlora
lora_model_dir:

sequence_len: 16384
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 3
optimizer: paged_adamw_32bit
# optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

loss_watchdog_threshold: 15.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 3
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

plugins:
- axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

outputs/qlora-out

This model is a fine-tuned version of pints-ai/1.5-Pints-16K-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9847

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.1396	0.0011	1	1.1313
1.0777	0.3332	295	1.0278
1.0219	0.6665	590	1.0119
1.0006	0.9997	885	1.0020
1.0385	1.3307	1180	0.9954
0.9405	1.6639	1475	0.9902
0.9249	1.9972	1770	0.9867
0.9951	2.3282	2065	0.9856
0.9713	2.6616	2360	0.9848
0.9576	2.9949	2655	0.9847

Framework versions

• PEFT 0.12.0
• Transformers 4.45.0.dev0
• Pytorch 2.4.1
• Datasets 2.21.0
• Tokenizers 0.19.1

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	4.66
IFEval (0-Shot)	17.24
BBH (3-Shot)	4.08
MATH Lvl 5 (4-Shot)	0.76
GPQA (0-shot)	0.00
MuSR (0-shot)	4.57
MMLU-PRO (5-shot)	1.30